Ayar Labs and Wiwynn have formed a strategic partnership to develop optically connected, rack-scale artificial intelligence (AI) systems designed to support the growing demands of hyperscale data centers. The collaboration combines Ayar Labs’ expertise in co-packaged optics (CPO) with Wiwynn’s capabilities in rack-level system design, integration, and high-volume manufacturing.
The initiative comes at a time when the rapid expansion of AI workloads is placing increasing pressure on traditional data center infrastructure. Copper interconnects, which have long served as the backbone of server connectivity, are reaching their practical limits in terms of bandwidth, power efficiency, and physical reach. Optical technologies are increasingly being explored as a solution to overcome these constraints and enable the next phase of AI infrastructure growth.
Through the partnership, the companies aim to create rack-scale architectures in which optical connectivity replaces conventional copper links. This approach is expected to improve bandwidth density, reduce power consumption, and support larger clusters of AI accelerators operating across multiple racks.
According to Mark Wade, AI infrastructure is evolving rapidly, requiring new interconnect technologies capable of supporting the scale and performance demanded by hyperscale computing environments. Optical interconnects integrated directly into system architectures, he noted, can eliminate bottlenecks and unlock significantly higher levels of efficiency and scalability.
The joint solution integrates Ayar Labs’ optical technologies—including its TeraPHY™ optical engines powered by the SuperNova™ remote light source—within Wiwynn’s rack-level infrastructure designed for next-generation AI data centers. The combined platform addresses several operational challenges faced by hyperscale operators, including optical fiber management, thermal design, integration of CPO-enabled AI application-specific integrated circuits (ASICs), and power optimization.
William Lin noted that silicon photonics is reshaping the way modern AI infrastructure is designed and deployed. By integrating optical I/O technologies with large-scale system architecture, the partnership seeks to accelerate the transition from research-driven photonics innovation to production-ready AI data center solutions.
The rack-scale infrastructure being developed through the collaboration is designed to support clusters of up to 1,024 AI accelerators or more. Each accelerator can deliver more than 100 terabits per second of optical connectivity, enabling thousands of processors to function as a unified computing system across multiple racks—an architecture increasingly required for large AI training models and complex inference workloads.
To support high-density computing environments, the platform incorporates a liquid-cooled system architecture optimized for high-power operation. The design also supports external laser small form factor pluggable (ELSFP) light sources, advanced fiber management systems, and serviceable rack configurations tailored for hyperscale data center deployments.
Wiwynn brings more than a decade of experience delivering rack-level IT infrastructure to global cloud service providers. The company has shipped servers and AI systems to more than 750 data centers worldwide and operates manufacturing facilities in Taiwan, the United States, Mexico, Malaysia, and the Czech Republic.
The companies plan to demonstrate the rack-scale AI solution at the Optical Fiber Communication Conference 2026, scheduled to take place March 15–19 in Los Angeles.
